gomp-discuss
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Gomp-discuss] OpenMP: Examples 1


From: Lars Segerlund
Subject: Re: [Gomp-discuss] OpenMP: Examples 1
Date: Fri, 27 Aug 2004 18:38:34 +0200

 There is a slight problem with the fortran runtime libs, ( which is quite easy 
to overcome but anyhow ), the standard specifies that the libs have to be 
thread safe ( I think they exclude IO ), which they are not. This can be 
handled later however.

 I was thinking of collecting some fortran examples and rip some parts out of 
the fortran spec. perhaps give the fortran binding module a go.

 Thus we would have a gomp.h for C and a module for fortran ready ( there is a 
prototype in the spec I jsut have to check it out a bit in conjunction with 
gfortran ).

 / Lars Segerlund.


On Fri, 27 Aug 2004 10:13:05 -0400
Scott Robert Ladd <address@hidden> wrote:

> Lars Segerlund wrote:
> >  The OdinMP compiler did compile for MPI as one of its targets.
> 
> Good!
> 
> I'm working up the basic for/do-loop examples in C and Fortran 95 today. 
>          I'll try to post those over the weekend.
> 
> ..Scott
> 
> > 
> >  / Seger.
> > 
> > 
> > On Thu, 26 Aug 2004 13:59:40 -0400
> > Scott Robert Ladd <address@hidden> wrote:
> > 
> > 
> >>Lars Segerlund wrote:
> >>
> >>> Scott, you are my hero :-) ...  don't give up !
> >>>
> >>> Excellent, stuff like this is what we need.
> >>
> >>Thanks for the support!
> >>
> >>
> >>> A question though, does it sound reasonable that we should go for support 
> >>> for both threads & MPI ? This would mainly concearn the lib(s), and 
> >>> perhaps we have to stay out of trouble during the implementation.
> >>
> >>A tricky question. OpenMP tends to be associated with low-latency 
> >>parallelism, while MPI and PVM represent various forms of message 
> >>passing mechanisms. In some cases, people have tried to write mixed MPI 
> >>and OpenMP programs, where the distributed nodes are multiprocessor 
> >>systems. One of the ASCI machines (I forget which color) uses 
> >>multiprocessor nodes connected via a high-speed network.
> >>
> >>I think we should just stick to low-latency parallelism in GOMP; 
> >>otherwise, I think we're mixing what some people see as apples and oranges.
> >>
> >>..Scott
> >>
> >>-- 
> >>Scott Robert Ladd
> >>Coyote Gulch Productions (http://www.coyotegulch.com)
> >>Software Invention for High-Performance Computing
> >>
> >>
> >>_______________________________________________
> >>Gomp-discuss mailing list
> >>address@hidden
> >>http://lists.nongnu.org/mailman/listinfo/gomp-discuss
> > 
> > 
> > 
> > _______________________________________________
> > Gomp-discuss mailing list
> > address@hidden
> > http://lists.nongnu.org/mailman/listinfo/gomp-discuss
> > 
> 
> 
> -- 
> Scott Robert Ladd
> Coyote Gulch Productions (http://www.coyotegulch.com)
> Software Invention for High-Performance Computing
> 
> 
> _______________________________________________
> Gomp-discuss mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/gomp-discuss




reply via email to

[Prev in Thread] Current Thread [Next in Thread]